The purpose of this study is to evaluate the applicability of 3D coordinate signals estimated for body parts using posture estimation AI from videos by comparing them with signals detected by a 3D motion analysis system.
The study involved one male (34 years old) and one female (58 years old), both unaffected in daily activities. Participants were informed of the research objectives and methods and provided consent. We conducted motion analysis of walking (2 cycles) and stand-up (1 movement). The left sagittal plane of the subject was recorded from the start to the end of the movement using a Panasonic LUMIX DMC-TZ70 camera at 240 fps. Simultaneously, the MAC3D System (NAC) captured 3D coordinates of body landmarks (both acromions, both trochanters, both fibular heads, both lateral malleoli) at 240fps. From the recorded video, 3D coordinate data of each body part was obtained using the posture estimation AI (Google MediaPipe Pose). Additionally, we computed the cross-correlation functions between these signals and the signals obtained from the 3D motion analysis system using the statistical software R. The peak values of the cross-correlation function that were statistically significant were used in the results.
For stand-up, the peak value of the cross-correlation function is significant and the signal is strongly correlated on all trials, the maximum value of the signal type and correlation coefficients were, respectively, left shoulder forward/backward (0.82), right shoulder forward/backward (0.79) left hip forward/backward (0.79), right hip forward/backward (0.81), left hip vertical (0.81), and right hip vertical (0.79). Similarly, in walking, they were left shoulder forward/backward (0.997), right shoulder forward/backward (0.997), left hip forward/backward (0.997), right hip forward/backward (0.997), left knee forward/backward (0.997), right knee forward/backward (0.995), left foot forward/backward (0.995) and right foot forward/backward (0.994).
This study confirmed the potential utility of motion analysis using pose estimation AI for the signals presented in the results. Moving forward, efforts will focus on improving the accuracy of all estimated signals.
This study is expected to facilitate quantitative evaluation of various movements in environments more closely related to daily life.
3D Coordinates signal
Posture Estimation
